Jailbreaking on HarmBench Visual Object Replacement
34.5ASR (K=1)Gemini 3 Flash
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini 3 Flash2026.05 | 34.5 | 52 | |
| Gemini 3.1 Pro2026.05 | 31 | 45.6 | |
| Qwen3-VL-235B2026.05 | 30.1 | 35.6 | |
| Qwen3-VL-32B2026.05 | 28.1 | 41.1 | |
| GPT-5.22026.05 | 6.8 | 11.5 | |
| Claude Haiku 4.52026.05 | 0.7 | 4.1 |